grison

grison turns raw security-scanner exports into clean markdown findings, gives you a plain-files workspace to triage and edit them in, and syncs the result into Ghostwriter (findings + evidence) and BookStack (methodology) with a git-style 3-way merge.

Markdown in the middle is the point: findings are files, so the transform layer is whatever edits files best — a human in any editor, a script, or an LLM let loose on the workspace. grison itself has no AI subsystem; its job is to validate what changed and sync it safely.

Supported scanner exports (auto-detected): Acunetix, Burp, Nessus, Nmap, OpenVAS, Qualys, sslyze, ZAP.

Install

pipx install grison            # or: uv tool install grison
grison --install-completion    # optional: shell tab completion (once, then restart shell)

For development: uv sync && uv run pytest (Python 3.11+, managed with uv).

Quickstart

mkdir engagement && cd engagement
grison parse scan.nessus burp.xml    # first run scaffolds the workspace
# triage findings/inbox/*.md, edit freely, move the keepers into place:
cp findings/inbox/sqli-login.md findings/reports/6-acme-q3/
grison status findings/              # validity report per record
grison sync                          # reconcile with Ghostwriter + BookStack

The three verbs

There is deliberately no pull/push (sync derives direction), no init (the first run of any verb scaffolds the workspace), and no validate (status reports, sync enforces). Moving a finding between tiers is a plain cp/mv.

The workspace

The directory tree is the data model — location is identity:

findings/
  inbox/
  library/
  reports/<id>-<slug>/
    narrative/
    evidence/
methodology/
  library/<book>/<chapter>/
  checklists/<engagement>/
.grison/

BookStack's structure mirrors losslessly: library/<book>/<page>.md for pages at a book's root, library/<book>/<chapter>/<page>.md for chaptered pages. Every book and chapter — including empty ones — materializes as a directory holding a .book.yml / .chapter.yml mirror (ids, name, description, shelf membership, chapter order; pull-only). Page sort order (priority) and page tags live in the page frontmatter and sync in both directions. Moving a file between chapter directories moves the page on BookStack; a page moved into a chapter remotely relocates the local file on the next sync.

Sync matches records by remote id stored in the file, not by filename (filenames are cosmetic). A file whose directory disagrees with its stored id is a move and becomes a new record at the destination.

Finding schema

One tier-agnostic schema: structured facts in YAML frontmatter, prose in fixed ## sections.

---
grison:
  kind: finding
  tier: instance            # library | instance
  gw: { table: reportedFinding, id: 183, report_id: 6 }
  synced: { hash: sha256:…, at: 2026-07-14T12:00:00Z }  # the 3-way merge base
severity: high              # informational|low|medium|high|critical
finding_type: web           # network|physical|wireless|web|mobile|cloud|host
cvss: { vector: "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H", score: 9.8 }
cwe: ["CWE-79"]             # validated against an embedded CWE index
affected_entities: |        # instances only
  https://app.example/
evidence:                   # instances only
  - { file: evidence/shell.png, caption: Shell, friendly_name: shell, gw: { id: 38, hash: sha256:… } }
---

# {title}
## Description
## Impact
## Mitigation
## Replication Steps
## References

The pydantic models in grison/model/ are the schema. CVSS is accepted as well-formed 3.0 or 3.1, as authored. CWE ids are validated against a vendored index of the MITRE catalog, so everything works fully offline.

How sync decides direction

Nobody chooses push or pull — the stored merge base (synced.hash) determines it per record:

• only local changed → push
• only remote changed → pull
• both changed → collision: the remote version is written to an x.remote.md sidecar, local is never overwritten; resolve, then grison sync --force-local <file> or --force-remote <file>
• both converged under a stale base → repair the base, write nothing

Guardrails

Guardrails stop anomalous or destructive outcomes, never routine ones — no confirmation prompts, --dry-run is opt-in. Three layers:

1. Validation — silent when green; a record must be valid to sync.
2. Trip-wires — fire only on anomaly: mass-change guard, structure drift, collision, duplicate identity, broken evidence link.
3. Snapshots — every remote write batch is snapshotted first, with a paired rollback.py, so every write is reversible.

Markdown ⇄ Ghostwriter HTML

Ghostwriter's rich-text fields use a small closed vocabulary, and the converter fails loudly on anything outside it rather than corrupt silently. Inline: **bold**, `code`, *em*, [links](…). Block: paragraphs and unordered/ordered lists (- and 1. , one level of 2-space sub-items, freely nested either way). Rejected inside a field: tables, images, headings (the ## section headers are grison structure mapping to Ghostwriter's separate fields, not field content). Constructs grison canonicalizes on purpose — editor highlight spans, non-standard link rel/target — are reported as sync warnings when dropped, never absorbed silently. BookStack pages are markdown-native and mirror verbatim, no conversion; pages authored in the WYSIWYG editor are skipped loudly until converted to markdown in BookStack.

What syncs, what doesn't

Every remote field has an explicit verdict — "not synced" always means grison can neither lose it nor damage it (partial updates never touch unlisted columns).

Two-way synced (in the merge base — edits on either side propagate, both-changed collides): finding title, severity, type, CVSS vector + score, all five prose sections, affected entities; tags and CWE ids (Ghostwriter's tag table, CWE:<n> convention); evidence images with caption / friendly name / description (per-image merge base); report narrative sections (report.extraFields), each section its own merge base; BookStack page body, name, priority, tags, and chapter/book location.

Mirrored read-only (pull-only; local edits are detected and rejected, never silently discarded): report metadata (.report.yml — project, client, dates, lifecycle flags); BookStack book/chapter/shelf structure (.book.yml, .chapter.yml, .shelves/*.yml — carry a READ-ONLY header; hand-edits error with a .remote.yml sidecar).

Deliberately not synced (remote-owned, unreachable by grison's writes): Ghostwriter report-finding position (per-severity kanban order, server renumbers), complete/assignedTo QA state, finding/report comments, Observations, report templates and reportConfiguration, report-level evidence rows (never deleted by grison), dead-in-corpus columns (hostDetectionTechniques, networkDetectionTechniques, findingGuidance, per-finding extraFields); BookStack container tags/covers/templates, page template flag, revision history. Renaming books/chapters happens in BookStack, not by renaming local dirs (a rename tripwire blocks the ambiguous case). Severity/finding-type id mappings are verified against the live instance at sync start and abort on drift.

Configuration

The first run in a directory scaffolds .grison/env — a commented template, chmod 600, auto-gitignored. Paste values there, or set the same keys as environment variables (env vars override the file; useful for CI):

Settings

Two more keys in the same file — non-secret behavior toggles, same env-var-overrides-file precedence:

License

MIT. Named after the mustelid.